AI ‐driven modelling and optimization of coal gasification in a 15MWth KEDA ® circulating fluidized bed: A CNN – DNN hybrid model with SHAP analysis

Abstract Fluidized bed coal gasification is a key technology for clean coal utilization, yet its inherent nonlinearity and multi‐physics coupling challenge traditional modelling approaches. Artificial intelligence (AI) offers a promising pathway to capture these complex relationships from industrial data, enabling accurate prediction and process optimization. Based on 2018 operational samples from a 15MWth KEDA® circulating fluidized bed coal gasifier, this study developed and compared four models for gas flow rate prediction: polynomial regression, random forest, DNN, and CNN–DNN hybrid. The CNN–DNN hybrid model achieved the best performance ( R 2 = 0.8004, MAPE = 0.5044%) with the most consistent residual variance, and while its accuracy gain over DNN ( R 2 = 0.7872) was modest, it reduced trainable parameters by 62.7% and learned relationships consistent with gasification principles. SHAP analysis identified coal feed rate (26.5%), top pressure (22.7%), and air flow rate (15.4%) as the three key parameters, and partial dependence plots informed an optimization strategy of ‘stabilizing coal feed, increasing top pressure, and raising air flow appropriately’. This work offers an interpretable data‐driven approach for soft sensing and operational optimization of coal gasification processes.